基于独立分量分析的遥感图像分类技术  

Remote Image Classification Based on Independent Component Analysis

作　　者：曾生根[1] 王小敏[1] 范瑞彬[1] 夏德深[1] 

机构地区：[1]南京理工大学计算机科学与工程系,江苏南京210094

出　　处：《遥感学报》2004年第2期150-157,共8页NATIONAL REMOTE SENSING BULLETIN

基　　金：南京市科委资金资助 (编号 993 11)

摘　　要：遥感图像的自动分类方法一般基于图像的统计信息。多光谱遥感图像之间有着一定的相关性 ,对遥感图像的自动分类有不利影响。一般用主成分分析去除波段之间的相关性。独立分量分析能利用相对主成分分析更高的统计分量 ,不但可以获得去相关的效果 ,而且可以得到相互独立的结果波段图像。本文首先讨论了独立分量分析的基本原理。在此基础上 ,介绍FastICA算法 ,并对其进行改进 ,得到M FastICA算法 ,并将其应用到遥感图像的分类上。实验结果表明 ,M FastICA算法较FastICA算法收敛性大为改善 。The automatic classification methods for remote sensing images are usually based on statistic information of the images. It has correlation among multi-spectral remote sensing images, and the correlation is a disadvantage to automatic classification of remote images. Commonly, Principal Component Analysis (PCA) is used to remove the correlation. Independent Component Analysis (ICA) can obtain higher order statistics information than PCA. It not only can remove the correlation, and also can obtain band images that are mutual independent. Firstly the fundamental of Independent Component Analysis is briefly introduced. Then, a fast algorithm of ICA (FastICA) and its modification (M-FastICA) are introduced, and are used to classify the remote sensing images. In the result, compare to basic FastICA algorithm, M-FastICA runs quickly and has better convergence performance, and improves the validity of the ICA in classifying of the remote sensing images.

关 键 词：独立分量分析 遥感图像 主成分分析 固定点算法 自适应最小距离分类法 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]